New AI Model Uses Complex Math to Rival Transformers on Language Tasks

AI Research Correspondent5d agoArXiv CS.CL✓Verified across 1 source

The Brief

Researchers introduced Phase-Associative Memory (PAM), a neural network using complex-valued math and matrix operations to model language sequences. PAM achieved 30.0 perplexity on WikiText-103, nearly matching transformer performance (27.1) while exploring whether quantum-like properties could improve language understanding.

✓Verified across 1 independent source

Sources

01https://arxiv.org/abs/2604.05030

New AI Model Uses Complex Math to Rival Transformers on Language Tasks

AI Models Play Cards Against Humanity — and Agree With Each Other More Than With Humans

Sam Altman's Home Targeted in Second Attack Within 48 Hours

LLMs Lose Ground to Lightweight Graph Parsers When Relation Extraction Gets Complex